ChatGPT in Slack and Microsoft Teams for Small Business (2026): Pricing, Permissions, and Setup

ChatGPT in Slack and Microsoft Teams is best treated as a company-managed assistant for approved conversations—not as a free chatbot bundled with either messaging platform. In channels, threads, and supported direct or personal chats, people can mention @ChatGPT to summarize the visible conversation, draft a reply, or use tools the ChatGPT admin has attached to that deployment. It uses the conversation context plus the deployment’s service account, plugins, and company-managed connections. It does not automatically inherit each participant’s private files, app permissions, ChatGPT memory, or private-channel access.

For a small business, the buying decision is mostly about identity and scope. OpenAI’s current pricing lists ChatGPT Business at $20 per user per month billed annually or $25 per user per month billed monthly, with a two-user minimum. Slack or Microsoft Teams licensing is a separate cost. OpenAI’s Slack and Teams admin guide says to confirm that @ChatGPT is available for your organization before setup; it does not publish a complete plan-by-plan entitlement matrix or promise that one paid ChatGPT seat covers unlimited messaging-platform users.

Pricing, plan, setup, permission, and surface claims were rechecked against current OpenAI, Slack, and Microsoft documentation on October 5, 2026. This is a vendor-document-verified editorial analysis, not a hands-on deployment test. Launch behavior can change.

The deployment combines conversation context with explicitly configured tools. It does not silently become every participant’s personal ChatGPT account.

What ChatGPT actually does inside Slack and Microsoft Teams

The new surface answers inside approved messaging conversations and can use approved tools without turning the whole workspace into one giant ChatGPT chat. OpenAI describes three core jobs: summarize information, draft responses, and use configured tools.

OpenAI warns that Slack and Teams can differ in conversation types and feature support. A buyer should test the exact channel, thread, and personal-chat pattern it plans to use instead of assuming feature parity.

The context model is intentionally narrower than many launch headlines imply. OpenAI says @ChatGPT uses configured connections and conversation context, but does not inherit a person’s file, app, or private-channel access. Shared replies are visible to the other participants in that conversation. At launch, the surface does not include ChatGPT memory or use a person’s saved ChatGPT memories.

That makes threads useful for bounded collaboration. A team can ask for a summary, decision list, draft response, or source-backed status update while keeping the relevant discussion visible. It does not make the bot an all-seeing company historian. If the needed information is in another system, an administrator must deliberately connect that system and grant the acting account access.

Which ChatGPT plans qualify, and does every participant need a seat?

For small businesses, ChatGPT Business is the relevant public managed-workspace plan, while Enterprise and Edu provide broader governance options; personal plans are not documented as the managed deployment path for this surface. However, OpenAI’s current Slack and Teams admin guide does not publish a definitive entitlement table. It tells administrators to confirm availability for their organization before setup.

A two-seat ChatGPT Business workspace therefore starts at $40 per month on annual billing or $50 per month on monthly billing, before tax, Slack or Teams fees, connected-app fees, and any additional usage or credit charges.

Does every Slack or Teams participant need a ChatGPT seat?

Not every participant needs to authenticate the shared company account, but OpenAI’s public admin guide does not clearly promise seat-free access for every messaging-platform participant. Members using a workspace connection do not each sign into that shared account. The surface’s service account and selected company connection perform the shared work.

That is different from saying “buy two Business seats and give unlimited Slack users ChatGPT.” Audience assignment, workspace membership, requesting-user permissions, the connected tool, and the feature being invoked can all matter. For example, Codex Cloud delegation from Slack requires access for both the Slack surface’s service account and the requesting users, even though the service account finds the environment and does not need GitHub access itself.

Budget conservatively: verify the intended participant group in the admin console and written terms before treating non-seat users as entitled. The sentence sales teams hate most is also the useful one: if the license matrix is not documented, it is not a discount yet.

ChatGPT, Slack, and Teams costs are separate layers

The @ChatGPT surface does not turn Slack or Microsoft Teams into a free entitlement. A business pays for the messaging platform it chooses, the qualifying ChatGPT workspace, and any third-party systems connected to the workflow.

Slack’s current US list pricing includes Free at $0, Pro at $8.75 per user monthly or $7.25 per user monthly on annual billing, Business+ at $18 monthly or $15 annually, and Enterprise+ at contract pricing. Slack Free permits up to 10 apps and keeps 90 days of searchable message history, which may be too restrictive for a governed pilot even when the app can be installed.

Microsoft sells Teams through standalone and Microsoft 365 options. Teams Essentials is aimed at small businesses, while Microsoft 365 Business plans combine Teams eligibility with varying combinations of Office apps, identity, email, storage, and administration. Pricing and Teams inclusion vary by market and plan, so use Microsoft’s live Teams comparison for the tenant you will deploy—not a copied US figure from an old article.

Service account, conversation context, plugins, and workspace connections are different

Four separate concepts determine what the assistant knows and can do. Mixing them together is how a pilot accidentally gets production authority.

Identity, tools, and data access are independent layers. A plugin can be available while its connected account remains narrowly scoped.

Conversation context

Conversation context is the channel, thread, direct message, or Teams chat content available to the request. It gives the model local meaning. It does not grant new access to private channels, personal drives, or unrelated chats.

The surface service account

The service account is the non-human ChatGPT workspace identity for the deployment. It can receive roles, groups, plugins, and connections, and its runs can be attributed in available analytics and audit records. It does not inherit the permissions of the administrator who created it.

This identity should be granted only the groups, plugins, and connections required for the pilot. OpenAI’s broader service-account documentation says unused accounts should be disabled or deleted and that either action revokes active tokens. The exact service-account features available depend on the workspace’s commercial configuration.

Plugins

A plugin packages reusable instructions and may include an MCP server that reads or acts on another system. Making a plugin available does not automatically grant access to that system. Administrators separately control plugin availability, MCP access, actions, approval behavior, and the external account’s scopes.

Personal app authorization

A personal plugin connection uses the individual’s own account and permissions. Where the messaging surface supports it, the person must authorize that access separately. Personal authorization should not be confused with the shared company connection, and it is a poor default for a repeatable company workflow because access can disappear when the employee leaves.

Company-managed workspace connections

A workspace connection lets the surface use an approved company account without every participant entering that account’s credentials. It may be a dedicated user, provider-specific service identity, or another supported authentication method. The external provider still controls the real data and action permissions.

Name connections by scope—such as “Support knowledge: read only”—instead of “Company Google.” Give each one a connection owner, permitted data list, approved actions, removal contact, and test case.

How admins set up ChatGPT in Slack or Teams

Setup requires both an OpenAI administrator and a messaging-platform administrator. One side cannot safely finish the deployment alone.

Slack setup

1. In the OpenAI admin console, select the intended ChatGPT workspace and open Agents.
2. Connect Slack and review the requested app permissions.
3. Have a Slack owner or authorized app manager approve the app and any additional permissions required by the new @ChatGPT experience.
4. Create one surface for the Slack workspace. Enterprise Grid requires a separate surface for each connected workspace.
5. For a pilot, choose Selected channels only and add one approved channel.
6. Test the channel and a direct message before attaching company data.

The channel allowlist does not block direct messages. If DMs are outside the pilot, treat that as an explicit policy and testing problem rather than assuming the channel setting solved it.

Microsoft Teams setup

1. Have the Teams administrator enable the app for the tenant and assign it to the intended users.
2. Obtain any required Microsoft administrator consent through the approved installation route.
3. In the OpenAI admin console, connect the same Teams tenant and create the corresponding ChatGPT surface.
4. Add the app to the approved team or supported chat location.
5. Test assigned and unassigned users and personal chat only where the surface supports it.

Add company access only after the empty-surface test

After the messaging layer works, add one approved plugin and one workspace connection. Verify the acting account, the records it can read, the destinations it can write to, the recipients who can open generated files, and the result of every enabled write or send action.

Read, write, approval, audit, spend, and offboarding boundaries

A safe rollout separates the ability to see data from the ability to change it. The useful control stack is layered: platform audience, surface scope, plugin availability, connection permissions, action controls, approval policy, and source-system authorization.

OpenAI’s Compliance API guidance is intended for supported security, legal, governance, and investigation records—not productivity scoring. Event coverage, retention, and access requirements can change, so the Admin API reference is the source of truth. Slack or Microsoft logs remain separate evidence. Do not assume one audit export contains the complete story across ChatGPT, the messaging platform, and every connected service.

For spend, ChatGPT Business list price is only the floor. Track accepted outputs, correction time, admin time, connected-system fees, and any usage credits. A bot that saves five minutes and creates twenty minutes of permission archaeology is not automation; it is a scavenger hunt with a logo.

Offboarding should remove access at every layer:

1. Remove the person or group from the surface audience.
2. Remove or reassign Slack/Teams app access and channel membership.
3. Revoke personal authorizations that are no longer appropriate.
4. Revoke or replace company-managed connections owned by a departing employee.
5. Disable or delete unused service identities and revoke their tokens.
6. Review scheduled or delegated work separately.
7. Retest approved and excluded conversations after the change.

What this is not

@ChatGPT is a messaging surface, not a synonym for every AI feature available in Slack, Teams, or ChatGPT.

Choose @ChatGPT when the work begins in a conversation and benefits from a visible shared response. Choose deterministic automation when a known event should always produce the same controlled action. Choose a custom bot when the business needs application-specific behavior and can own the engineering. Choose native Slack or Microsoft AI when the source of truth and desired governance already live primarily in that suite.

For adjacent buying decisions, see ChatGPT Business vs Microsoft 365 Copilot, ChatGPT Work for small business, and best automation software for small business.

A least-privilege 14-day pilot

Start with one internal channel, one read-only source, explicit mentions, and no write actions. A low-risk first workflow is a weekly internal policy-and-project FAQ in a private operations channel.

Prove read boundaries and answer quality before enabling any write action.

Days 1–2: define the boundary

• Name one ChatGPT admin, one Slack or Teams admin, one connection owner, and one pilot owner.
• Use approved, non-sensitive test documents with no customer, payroll, health, legal, or payment data.
• Define the allowed output: answer a question, cite the source, and say when evidence is missing.
• Define stop conditions: excluded data appears, an unsupported claim lacks a source, or any write action is attempted.

Days 3–5: test the empty surface

• Install the app in one private pilot channel or team.
• Require an explicit @ChatGPT mention.
• Test an approved channel, excluded channel, thread, direct message or personal chat, assigned user, and unassigned user as applicable.
• Confirm who can see replies and open generated links.

Days 6–10: attach one read-only connection

• Use a dedicated company-managed account with access only to the pilot folder or knowledge source.
• Enable only the required read action.
• Run 20 representative questions, including ambiguous, stale, missing, and permission-denied cases.
• Score source accuracy, unsupported statements, correction time, and accidental exposure.

Days 11–14: decide without adding authority

• Review available activity and platform logs.
• Confirm the monthly cost floor and any usage or connected-app charges.
• Keep writes disabled unless a separate, reversible action has a clear owner and approval step.
• Expand only if the surface stayed inside scope and at least 80% of answers were usable with minor edits.

Suggested stop criteria are zero tolerance for unauthorized access or writes, and a redesign threshold if more than 20% of answers require substantial correction. These are operating recommendations, not OpenAI performance guarantees.

The small-business verdict

ChatGPT in Slack and Microsoft Teams is worth piloting when a team already works in one messaging platform, needs shared answers or drafts, and can assign real owners to permissions and connections. Its advantage is shared conversational access to approved tools. Its risk is that the shared identity may reach data participants could not access individually.

Proceed if the business can isolate one low-risk workflow, confirm eligibility in the admin console, use a dedicated read-only account, and test channels, DMs or chats, participants, and exclusions. Wait if the team expects one cheap seat to cover everyone, needs a published feature-parity guarantee, cannot audit connected actions, or wants autonomous writes on day one.

FAQ

Is ChatGPT free inside Slack or Microsoft Teams?

No. Slack or Teams is a separate platform cost, and the company needs an eligible managed ChatGPT workspace. Connected business systems may also charge separately.

Does every participant need a ChatGPT Business seat?

Participants do not each authenticate the shared company account, but OpenAI’s public admin guide does not promise unlimited seat-free participation. Confirm the intended audience and licensing in the admin console and contract before budgeting.

Can ChatGPT read every Slack channel or Teams conversation?

No. It uses the current conversation context and the access configured for the surface and its connections. Slack surfaces can be limited to selected channels, although that setting does not block direct messages.

Does it use my personal ChatGPT memory?

No at launch. OpenAI says @ChatGPT does not include memory or use a person’s ChatGPT memories.

Can it update a CRM or send a message?

Only when an approved plugin, connection, action permission, and source-system identity allow that action. Start read-only and test every write or send path separately.

Is this the same as Microsoft 365 Copilot or Slackbot?

No. Those are native platform assistants with their own data boundaries, licensing, and controls. @ChatGPT is an OpenAI-managed surface inside the messaging platform.

Was this guide based on hands-on testing?

No. It is based on current vendor documentation. Availability, conversation types, action support, plan eligibility, and audit coverage should be reverified immediately before deployment.